PrefectHQ/fastmcp · error · ValueError

Unsupported tool result content type: {type(item).__name__}

Error message

Unsupported tool result content type: {type(item).__name__}

What it means

_sampling_content_to_google_genai_part raises ValueError when a ToolResultContent block contains an item whose type is not TextContent (e.g. ImageContent or EmbeddedResource). Google GenAI function responses are text-only here, so non-text tool-result items are rejected.

Source

Thrown at fastmcp_slim/fastmcp/client/sampling/handlers/google_genai.py:249

        # See: https://ai.google.dev/gemini-api/docs/thought-signatures
        return Part(
            function_call=FunctionCall(
                name=content.name,
                args=content.input,
            ),
            thought_signature=b"skip_thought_signature_validator",
        )

    if isinstance(content, ToolResultContent):
        # Extract text from tool result content
        result_parts: list[str] = []
        if content.content:
            for item in content.content:
                if isinstance(item, TextContent):
                    result_parts.append(item.text)
                else:
                    msg = f"Unsupported tool result content type: {type(item).__name__}"
                    raise ValueError(msg)
        result_text = "".join(result_parts)

        # Extract function name from toolUseId
        # Our IDs are formatted as "{function_name}_{uuid8}", so extract the name.
        # Note: This is a limitation of MCP's ToolResultContent which only carries
        # toolUseId, while Google's FunctionResponse requires the function name.
        tool_use_id = content.tool_use_id
        if "_" in tool_use_id:
            # Split and rejoin all but the last part (the UUID suffix)
            parts = tool_use_id.rsplit("_", 1)
            function_name = parts[0]
        else:
            # Fallback: use the full ID as the name
            function_name = tool_use_id

        return Part(
            function_response=FunctionResponse(
                name=function_name,

View on GitHub (pinned to 1f02114297)

Solutions

1. Make tools return plain text results when the client uses the Gemini sampling handler.
2. Strip/downgrade non-text items from tool result content before invoking the handler.
3. Catch ValueError and serialize the offending item to a text description in your sampling callback.
4. Use a handler/provider that supports multimodal tool results if rich results are required.

Example fix

// before
def my_tool(path: str) -> ImageContent: ...
// after
def my_tool(path: str) -> str:
    return f"Image at {path}: <describe or omit>"

from mcp.types import TextContent
def validate_tool_results(messages):
    for m in messages:
        contents = m.content if isinstance(m.content, list) else [m.content]
        for c in contents:
            if hasattr(c, "content") and c.content:  # ToolResultContent
                for item in c.content:
                    if not isinstance(item, TextContent):
                        raise ValueError(f"Tool result item {type(item).__name__} not supported by Gemini handler")

def is_text_tool_result(c) -> bool:
    return not hasattr(c, "content") or all(isinstance(i, TextContent) for i in (c.content or []))

try:
    result = await handler(messages, params, context)
except ValueError as e:
    if "Unsupported tool result content type" in str(e):
        result = CreateMessageResult(content=TextContent(type="text", text="Tool returned non-text output that cannot be forwarded."), role="assistant", model="unknown")
    else:
        raise

When it happens

Trigger: A server returns a tool result whose content list includes images/audio/resources, and the client routes sampling through GoogleGenaiSamplingHandler.

Common situations: Tools returning screenshots or files in results; servers attaching resources to tool outputs; rich tool results designed for clients that support multimodal content.

Related errors

• Unsupported content type: {type(content)}
• Unsupported content type: {type(content)}
• Invalid message role: {message.role}
• No candidate in response from completion.
• Model returned only thinking/reasoning content with no respo

AI-assisted analysis of PrefectHQ/fastmcp@1f02114297 (2026-08-29). Data as JSON: /api/errors/9778528e5fd21dce. Report an issue: GitHub.